Smoking has been consistently associated with an increased risk of colorectal adenomas and hyperplastic polyps as well as colorectal cancer. Conversely, nonsteroidal anti-inflammatory drugs (NSAID) have been associated with reduced colorectal cancer risk. We conducted a population-based case-control study to evaluate the joint association between smoking and regular NSAID use with colorectal cancer risk; we also examined these associations stratified by tumor microsatellite instability (MSI). We analyzed 1,792 incident colorectal cancer cases and 1,501 population controls in the Seattle, Washington area from 1998-2002. MSI, defined as MSI high (MSI-H) or MSI-low/microsatellite stable (MSI-L/MSS), was assessed in tumors of 1,202 cases. Compared with nonsmokers, colorectal cancer risk was modestly increased among individuals who had ever smoked. Current NSAID use was associated with a 30% lower risk compared with nonusers. There was a statistically significant interaction between smoking duration and use of NSAIDs (Pinteraction = 0.05): relative to current NSAID users who never smoked, individuals who had both smoked for >40 years and had never used NSAIDs were at the highest risk for colorectal cancer (adjusted odds ratio, 2.8; 95% confidence intervals, 1.8-4.1). Compared with nonsmokers, there was a stronger association within MSI-H tumors with current smoking than there was within MSI-L/MSS tumors. Smokers of long duration were at elevated risk of MSI-H tumors even with NSAID use. The risk of MSI-L/MSS tumors was not elevated among long-duration smokers with long exposure to NSAIDs but was elevated among long-duration smokers who had never used NSAIDs. There seems to be a synergistic inverse association (implying protection) against colorectal cancer overall as a result of NSAID use and nonsmoking, but risk of MSI-H colorectal cancer remains elevated among smokers even when they have a history of NSAID use. (Cancer Res 2006; 66(13): 6877-83)
- colorectal cancer
- microsatellite instability
Both smoking history and nonsteroidal anti-inflammatory drugs (NSAID), including aspirin, have been extensively studied as risk factors for colorectal cancer. Smoking has been consistently associated with a two- to three-fold increased risk of colorectal adenomas ( 1) and has also been associated with a modestly increased risk of colorectal cancer ( 2). Although studies completed before 1970 found no association between cigarette smoking and colorectal cancer, more recent studies examining smoking history have observed an increased risk of colorectal cancer with long-term use of cigarettes ( 3– 15). Giovannucci et al. hypothesized that, after the initiation of the carcinogenic process by smoking, an induction period of several decades may be necessary before an association with colorectal cancer is observed ( 2– 4). Among the studies finding a positive association, some found that the smoking-associated elevation in risk of colorectal cancer persisted even among those who had quit >10 years before the study, and others reported that the amount smoked may be more important than the total number of years smoked ( 8– 10). Cigarette smoke contains hundreds of carcinogenic metabolic products, including nitrosamines, heterocyclic amines, and polycyclic aromatic hydrocarbons, which damage DNA ( 16); this accumulated damage may not be preventable or reversible.
There is also evidence that cigarette smoking may be more strongly associated with a subset of colorectal tumors displaying microsatellite instability (MSI; ref. 17), an acquired genetic characteristic of some tumors that is the result of defects in the DNA mismatch-repair machinery ( 18). MSI high (MSI-H) has been associated with other lifestyle and tumor characteristics ( 17, 19– 22). A strong association between smoking and hyperplastic polyps has also been observed ( 23); as a result, we hypothesized that, because smoking has been associated with MSI-positive tumors ( 17, 21), hyperplastic polyps may be precursor lesions in a pathway leading to MSI-positive tumors ( 23).
Cohort and case-control studies have consistently found that long-term users of aspirin or other NSAIDs have ∼30% lower risk of colorectal adenomatous polyps and colorectal cancer than nonusers ( 24– 36). Recently, two randomized controlled trials showed that the risk of recurrent adenomatous polyps was reduced by regular aspirin use; the reduction was as high as 41% after 1 year ( 37, 38). Widely proposed as chemopreventive agents, NSAIDs act, at least in part, by inhibiting cyclooxygenase (COX)-2, thus inhibiting prostaglandin synthesis ( 39). To our knowledge, there have been no published studies examining the relationship of NSAID use and colorectal cancer or polyps exhibiting MSI.
The detrimental effects of smoking seem to be durable and possibly irreversible, which makes the determination of whether the chemoprotective effects of NSAIDs extend to individuals with a history of smoking of considerable importance. We conducted a population-based case-control study to investigate the relationship between smoking and colorectal cancer incidence stratified by NSAID use; we also evaluated whether the relationship differed by the MSI status of the tumor.
Materials and Methods
Study subjects. Subjects for this analysis included cases and controls enrolled in the Seattle Colorectal Cancer Family Registry (Seattle CCFR), a member of the CCFR, a multisite international consortium ( 40). Eligible cases were identified through the Puget Sound Surveillance, Epidemiology, and End Results (SEER) Program, a population-based registry. The inclusion criteria for cases specified men and women residing in the Seattle metropolitan area (King, Pierce, and Snohomish Counties), ages 20 to 74 years, and diagnosed with incident colon or rectal cancer (ICD for Oncology, Ninth Edition C18.0, C18.2-18.9, C19.9, and C20.0-20.9; ref. 41) between April 1, 1997 and March 31, 2001.
After the cases were identified by the SEER Registry, their physicians were contacted about their eligibility for this study. If the physician did not object to the patient's participation, the individual was approached with an introductory letter and then by a telephone call. A total of 2,573 eligible cases were identified. Of these, physicians refused contact for 123 (5%), 343 (13%) had died or could not be located, and 276 (11%) declined to participate, for a final case participation proportion of 71% (n = 1,831).
Community controls were randomly selected to match the distribution of the cases on age and sex. The controls were selected from the same three Seattle metropolitan area counties as the cases. Individuals ages 20 to 64 years were selected from Washington state driver's license files; individuals ages >65 were selected from Health Care Financing Administration files. Phone numbers from eligible subjects were then obtained from published directories on CD-ROM or from professional tracing services. Cases and controls were limited to English-speaking individuals with available phone numbers where they could be contacted. Of the 2,363 potential controls identified, 162 (7%) had died or could not be located, and 666 (28%) declined to participate, for a final control participation proportion of 65% (n = 1,535).
Data. A structured telephone interview was used to collect data on smoking history, aspirin and other NSAID use, postmenopausal hormone (PMH) use by women, colorectal cancer screening history, family history of colorectal cancer, body size, demographics, and other lifestyle factors. Information collected about smoking history included whether the subjects had ever smoked cigarettes regularly, defined as smoking at least one cigarette daily for >3 months; when they first started smoking; the number of cigarettes typically smoked daily; whether they were current smokers; when they permanently stopped smoking; and total duration of smoking. Information was also gathered on whether aspirin or other NSAIDs were used regularly, defined as at least twice weekly for >1 month; how often they were taken; whether they were taken currently; and how long they have been used regularly.
MSI was assessed in tumors from 1,202 cases using nine markers: four mononucleotide repeats (BAT25, BAT26, BAT40, and BAT34C4), four dinucleotide repeats (ACTC, D5S346, D18S55, and D10197), and one complex marker (MYCL). These markers include the five recommended markers in the panel proposed during the National Cancer Institute (NCI) workshop on MSI for cancer detection ( 42). The PCR fragments were tagged with a fluorescent dye and analyzed on an ABI3100 genetic analyzer, using a protocol described previously ( 43). All MSI results were corroborated with immunohistochemical testing of hMLH1, hMSH2, and hMSH6.
Statistical analysis. Pack-years of smoking were calculated as the number of years smoked multiplied by the usual number of packs of cigarettes smoked daily; the reference category consisted of those who had no smoking history. Current users of NSAIDs were those who had taken either aspirin or other NSAIDs regularly about 2 years ago. Nonusers of NSAIDs use included only those who had used neither aspirin nor other NSAIDs regularly.
Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters (kg/m2). Based on the distributions of the male and female control populations, BMI was divided into quartiles for each sex. PMH use among postmenopausal women was categorized as never users, past users, and current users. Users of PMHs were considered those who had taken PMH for at least 6 months, and current users were those who were still taking PMH 2 years ago. Having a history of screening was defined as having had an endoscopy at least 2 years before the diagnosis, for those diagnosed with distal tumors. Tumors exhibiting MSI at >30% markers were classified as MSI-H, and those with MSI at <30% positive markers were classified as MSI-low/microsatellite stable (MSI-L/MSS).
Unconditional logistic regression was used to estimate the odds ratios (OR) and 95% confidence intervals (95% CI; ref. 44). Multivariable logistic models were used to adjust for potential confounders, including age at interview, BMI, sex, PMH use in women, family history, and history of screening. Stratified analyses were done to assess the interaction between NSAID use and smoking history, where the common reference group was those who both were never smokers and had the highest usage of NSAIDs. Log-likelihood ratio tests were calculated to test for significance of multiplicative interactions, comparing models that differed only by the cross-product term of smoking and NSAID use. Tests for trend were evaluated by including the predictor as a continuous variable in the regression model for both smoking and NSAID- use history. For NSAID use, tests for trend included all categories of use, but for smoking history, tests of trend did not include those who had never smoked. Subjects with missing data for either primary exposure (smoking or NSAID use) or who did not complete the questionnaire (39 cases and 34 controls) were excluded; thus, 1,792 cases and 1,501 controls were included in the analysis. For the MSI analysis, polytomous logistic regression models were used to compare MSI-H cases and MSI-L/MSS cases with controls, adjusting for the other covariates of interest as described above. All statistical analyses were done using SAS version 8.2 (SAS Institute, Inc., Cary, NC). All statistical tests were two sided.
Selected characteristics of cases and controls are presented in Table 1 . The mean age for cases was 60.5 years (SD, 10.3); for controls, the mean age was 61.4 years (SD, 10.1). In both men and women, the proportion of cases in the highest BMI category was much higher than the proportion of controls in the highest BMI category. Among women, cases were less likely to have ever used, and to currently use, PMH compared with controls. Approximately 16% of cases had a first-degree family history of colorectal cancer compared with 10% of controls.
Smoking was somewhat more common among cases than controls ( Table 2 ). Eighteen percent of cases and 13% of controls were current smokers, and 45% of cases and 43% of controls were former smokers. After adjustment, current smokers had a modestly increased risk of colorectal cancer ( Table 2), and there was a dose response, with increasing duration of smoking associated with increasing risk (Ptrend = 0.007). Similar results were observed with pack-years of smoking.
NSAID use was more common among controls than cases ( Table 2). Overall, 21% of cases and 24% of controls had used NSAIDs formerly, and 30% of cases and 34% of controls were current users. After adjustment, compared with nonusers, those who currently used NSAIDs had a reduction in colorectal cancer risk of 30% (95% CI, 0.6-0.9), and those who used NSAIDs for ≥10 years had a reduction in risk of about 40% (95% CI, 0.5-0.8).
The joint effects of smoking and NSAIDs on colorectal cancer risk are presented in Table 3 . Relative to current users of NSAIDs who were never smokers, smokers of >40 years who had never used NSAIDS had the highest risk (OR, 2.8; 95% CI, 1.8-4.1). There was a statistically significant interaction between smoking duration and history of NSAIDs use (P = 0.049). Although not statistically significant (P = 0.059), there was also a suggestion of a similar interaction between smoking duration and NSAID-use duration: colorectal cancer risk increased with increasing years of smoking and decreasing years of NSAID use. There were no statistically significant interactions between NSAID use and pack-years of smoking (results not shown).
When the cases were stratified by MSI status, there were stronger associations between smoking and colorectal cancer risk among individuals with MSI-H tumors than with MSI-L/MSS tumors: current smokers and smokers of >40 years were 2.6 to 2.9 times more likely to have MSI-H colorectal cancers than nonsmokers; the risk of MSI-L/MSS colorectal cancer was, at most, 1.5 times that of nonsmokers ( Table 4 ). When NSAIDs had been used for ≥10 years, there was a greater reduction in risk for MSI-L/MSS (40%), and no inverse association was seen for MSI-H colorectal cancers. Those who had any duration of smoking for >20 years were at an approximately two-fold increased risk of MSI-H colorectal cancers compared with nonsmokers even when they were current users of NSAIDs, with the highest risk among those who had smoked for >40 years and who had never used NSAIDs (OR, 4.5; 95% CI, 2.1-9.5; Table 5 ). This relationship was not observed in those with MSI-L/MSS colorectal cancers: risk was elevated only among those who smoked for >20 years and who had never used NSAIDs.
In this large population-based study, the risk of colorectal cancer was elevated in those who were current smokers and who had never used NSAIDs. Those who had both smoked and never used NSAIDs were at highest risk. Individuals who were current or former users of NSAIDs who had smoked for long durations were still at higher risk than nonsmokers, although there was evidence that the colorectal cancer risk associated with smoking may be reduced with NSAID use of ≥5 years. There was a dose response, with longer duration of smoking being associated with the least benefit of NSAID use. When stratifying by MSI status, the joint association between smoking and NSAID use in colorectal cancer risk for MSI-L/MSS tumors was similar to that of the combined group. However, there was an elevated risk of MSI-H colorectal cancer among smokers regardless of duration, even if they were currently using NSAIDs.
The 30% reduction in colorectal cancer risk found with current NSAID use is consistent with the range found in other studies ( 24– 32). The estimates we found for the risk of colorectal cancer from cigarette smoking are also within the range of estimates observed in previous studies ( 3– 15). These studies found relative risks between 1.4 and 3.1, with higher relative risks for those whose duration of smoking was longer (many decades) and those who smoked more frequently (higher number of packs of cigarettes daily). Our results are also consistent with several studies, which reported that greater duration of smoking was associated with increased risk of MSI-H colorectal tumors compared with that of MSI-L/MSS tumors ( 17, 21, 45). To our knowledge, no previous study has reported interactions between smoking history and NSAID use.
The large population-based design of the study, the comprehensive assessment of smoking and NSAID use history, and the collection of tumor-tissue samples from cases make this a powerful resource with which to evaluate the role of these exposures in colorectal cancer etiology. However, some limitations should be considered. Although the study included many participants and substantial numbers of long-term smokers and NSAID users, stratified analyses resulted in small-cell sizes, and power was limited in some strata. This may account for some of the inconsistencies for the interaction between smoking and NSAID use when stratified by MSI status. Recall bias may be a concern: there may be differential accuracy in reporting NSAID and smoking history between cases and controls, although it is not obvious which group would be likely to be more accurate. Selection bias in cases should be minimal because we attempted to enroll all incident cases of colorectal cancer that were reported to the population-based SEER Registry, and the response rate was good, with ∼65% to 70% of both cases and controls eligible participating in the study. Although we did not have tissue specimens on all cases, there were no significant differences in either potential confounders or risk factors of interest between individuals who had MSI status assessed and those who did not. A polytomous regression model was chosen to compare, simultaneously, each case group (defined by MSI status) to the controls, a preferred option over separate comparisons with controls or simple case-case analysis.
Because NSAIDs may reduce the risk of colorectal cancer by ∼30% to 60% ( 24– 35) and several studies have reported that up to one in five cases of colorectal cancer may be attributable to smoking ( 3, 5, 8, 9), it is important to know whether NSAIDs are associated with a reduction in risk in the presence of a smoking history. The results of our study suggest that NSAID use may not reduce the increased risk of colorectal cancer caused by smoking. In addition, it is important to consider the effects of smoking and NSAID use in those who have mutations in, or methylation of, genes that cause microsatellite-unstable tumors.
It is biologically plausible that cigarette smoke and NSAIDs may interact to affect colorectal cancer risk. Cigarette smoke has been well established as a carcinogen, and its effects on many organ sites are well established. It has been hypothesized that there are two steps by which smoking may lead to carcinogenesis (reviewed in ref. 46): initiation of damage to DNA by polycyclicaromatic hydrocarbons and promotion of initiated cells by carcinogens and other compounds. A lengthy induction period of several decades may be necessary, however, before an association, specifically with colorectal cancer, is observed ( 2– 4). Given the damage that smokers receive over their lifetime, even strong antiprogression agents, such as NSAIDs, may be ineffective. Indeed, former smokers have an increased risk of most cancers, including lung, esophageal, bladder, and pancreatic, even many years after cessation ( 47). It is plausible, as we observed, that cigarette smoke may be more strongly associated with MSI-H tumors. Although nucleotide excision repair will correct most DNA damage ( 48, 49), incomplete repair can result in miscoding ( 50). Ultimately, rapid or high level accumulation of such miscodings may overwhelm the mismatch-repair system, resulting in a tumor characterized by MSI ( 17). However, it also appears from our results and other reports ( 3, 4, 8, 12) that duration of smoking for several decades or more increases colorectal cancer risk, regardless of MSI status.
There are many potential mechanisms by which NSAIDs may reduce risk of colorectal cancer. The most commonly described mechanism involves the ability of NSAIDs to inhibit COX activity, thereby depressing prostaglandin synthesis ( 39, 51). Although the mechanism is not entirely understood, it has been shown that elevated levels of prostaglandins are associated with colon cancer through the stimulation of cell proliferation ( 52). NSAIDs, however, may not be able to counteract the long-term effects of smoking, as evidenced by our observation that long-term smokers are at increased risk of colorectal cancer despite current NSAID use.
The results of this study and of other studies examining differences between MSI-H and MSI-L/MSS colorectal cancer ( 17, 21, 45) and differences between adenomatous and hyperplastic polyps ( 23, 53) support at least two pathways to colorectal cancer: one is through the adenoma-carcinoma sequence ( 54) and the other one is through a hyperplastic polyp-serrated adenoma progression ( 23, 55, 56). Risk factors, particularly smoking and NSAID use, may affect these two pathways differently.
If the benefits of NSAIDs are indeed limited to nonsmokers, it will be important to examine the associations of NSAID use and smoking history with precursor lesions and adenomatous and hyperplastic polyps, as smoking is more strongly associated with increased risk of hyperplastic polyps ( 3, 4, 23, 57), and the protective effect of NSAIDs is probably more marked against adenomas ( 34, 35). It will also be useful to examine whether metabolic genetic variation further modifies these relationships, as we have recently observed that specific polymorphisms in the CYP2C9 and UGT1A6 genes obviate the protective effect of regular NSAID use ( 36).
Grant support: NCI grant U01CA74794 and NIH training grant 1R25CA094880. NCI, NIH grant CA-95-011 and through cooperative agreements with members of the Colon Cancer Family Registry and Principal Investigators.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
We thank Sushma S. Thomas, Christine F. Rimorin, and Amy French for technical assistance with MSI testing, Brenda Diergaarde for MSI interpretation, Seattle CCFR (U01 CA074794) for providing the data for the analysis, and all the CCFR investigators for their advice and feedback throughout the writing of this article.
- Received March 28, 2005.
- Revision received April 19, 2006.
- Accepted April 26, 2006.
- ©2006 American Association for Cancer Research.